524 37 14MB
English Pages [307] Year 2021
HANDBOOK OF DEEP LEARNING IN BIOMEDICAL ENGINEERING Techniques and Applications EDITED BY VALENTINA EMILIA BALAS Full Professor, Department of Automatics and Applied Software Aurel Vlaicu University of Arad, Romania
BROJO KISHORE MISHRA Professor, Department of CSE, School of Engineering GIET University, India
RAGHVENDRA KUMAR Associate Professor, Department of Computer Science and Engineering, School of Engineering GIET University, India
Academic Press is an imprint of Elsevier 125 London Wall, London EC2Y 5AS, United Kingdom 525 B Street, Suite 1650, San Diego, CA 92101, United States 50 Hampshire Street, 5th Floor, Cambridge, MA 02139, United States The Boulevard, Langford Lane, Kidlington, Oxford OX5 1GB, United Kingdom Copyright © 2021 Elsevier Inc. All rights reserved. No part of this publication may be reproduced or transmitted in any form or by any means, electronic or mechanical, including photocopying, recording, or any information storage and retrieval system, without permission in writing from the publisher. Details on how to seek permission, further information about the Publisher’s permissions policies and our arrangements with organizations such as the Copyright Clearance Center and the Copyright Licensing Agency, can be found at our website: www.elsevier.com/permissions. This book and the individual contributions contained in it are protected under copyright by the Publisher (other than as may be noted herein).
Notices Knowledge and best practice in this field are constantly changing. As new research and experience broaden our understanding, changes in research methods, professional practices, or medical treatment may become necessary. Practitioners and researchers must always rely on their own experience and knowledge in evaluating and using any information, methods, compounds, or experiments described herein. In using such information or methods they should be mindful of their own safety and the safety of others, including parties for whom they have a professional responsibility. To the fullest extent of the law, neither the Publisher nor the authors, contributors, or editors, assume any liability for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions, or ideas contained in the material herein. Library of Congress Cataloging-in-Publication Data A catalog record for this book is available from the Library of Congress British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library ISBN: 978-0-12-823014-5
For information on all Academic Press publications visit our website at https://www.elsevier.com/books-and-journals
Publisher: Mara Conner Acquisitions Editor: Chris Katsaropoulos Editorial Project Manager: Fernanda A. Oliveira Production Project Manager: Niranjan Bhaskaran Cover Designer: Matthew Limbert
Typeset by TNQ Technologies
Contributors Rashmi Agrawal Manav Rachna International Institute of Research and Studies (MRIIRS), Faridabad, Haryana, India
B. Balamurugan School of Computing Science and Technology, Galgotias University, Greater Noida, Uttar Pradesh
Aradhana Behura Veer Surendra Sai University of Technology, Burla, Sambalpur, Odisha, India
Saakshi Bhargava Department of Physical Sciences and Engineering, Banasthali Vidyapith, Tonk, Rajasthan, India
Son Dao International University, Vietnam National University - Ho Chi Minh City, Ho Chi Minh City, Vietnam
Jayashankar Das Centre for Genomics and Biomedical Informatics, Institute of Medical Sciences and SUM Hospital, Siksha ‘O’ Anusandhan Deemed to be University, Bhubaneswar, Odisha, India
R. Indrakumari School of Computing Science and Technology, Galgotias University, Greater Noida, Uttar Pradesh
Aashna Jha Department of Electronics and Communication, Netaji Subhas University of Technology, New Delhi, India
Vaishali Kalra The NorthCap University, Gurugram, Haryana, India
Supriya Khaitan School of Computing Science and Technology, Galgotias University, Greater Noida, Uttar Pradesh
M. Nagoor Meeral PG & Research Department of Computer Science, Sadakathullah Appa College, Tirunelveli, Tamil Nadu, India
Subhashree Mohapatra Department of Computer Science and Engineering, Institute of Technical Education and Research, Siksha ‘O’ Anusandhan Deemed to be University, Bhubaneswar, Odisha, India
S. Shajun Nisha PG & Research Department of Computer Science, Sadakathullah Appa College, Tirunelveli, Tamil Nadu, India
xii
Contributors
Tan Pham International University, Vietnam National University - Ho Chi Minh City, Ho Chi Minh City, Vietnam
T. Poongodi School of Computing Science and Technology, Galgotias University, Greater Noida, Uttar Pradesh
Shrddha Sagar School of Computing Science and Technology, Galgotias University, Greater Noida, Uttar Pradesh
Deepak Kumar Sharma Department of Information Technology, Netaji Subhas University of Technology, New Delhi, India
Srishti Sharma The NorthCap University, Gurugram, Haryana, India
Pawan Singh Department of Information Technology, Netaji Subhas University of Technology, New Delhi, India
G. Sudha Sadasiuvam Department of CSE, PSG College of Technology, Coimbatore, Tamil Nadu, India
M. Mohamed Sathik PG & Research Department of Computer Science, Sadakathullah Appa College, Tirunelveli, Tamil Nadu, India
Tripti Swarnkar Department of Computer Application, Institute of Technical Education and Research, Siksha ‘O’ Anusandhan Deemed to be University, Bhubaneswar, Odisha, India
Siddharth Verma Manav Rachna International Institute of Research and Studies (MRIIRS), Manav Rachna Campus, Faridabad, Haryana, India
About the editors Valentina Emilia Balas, Brojo Kishore Mishra, Raghvendra Kumar
Valentina E. Balas is currently Full Professor in the Department of Automatics and Applied Software at the Faculty of Engineering, “Aurel Vlaicu” University of Arad, Romania. She holds a Ph.D. Cum Laude, in Applied Electronics and Telecommunications from Polytechnic University of Timisoara. Dr. Balas is author of more than 350 research papers in refereed journals and International Conferences. Her research interests are in Intelligent Systems, Fuzzy Control, Soft Computing, Smart Sensors, Information Fusion, Modeling and Simulation. She is the Editor-in Chief to International Journal of Advanced Intelligence Paradigms (IJAIP) and to International Journal of Computational Systems Engineering (IJCSysE), member in Editorial Board member of several national and international journals and is evaluator expert for national, international projects and PhD Thesis. Dr. Balas is the director of Intelligent Systems Research Centre in Aurel Vlaicu University of Arad and Director of the Department of International Relations, Programs and Projects in the same university. She served as General Chair of the International Workshop Soft Computing and Applications (SOFA) in nine editions organized in the interval 2005-2020 and held in Romania and Hungary. Dr. Balas participated in many international conferences as Organizer, Honorary Chair, Session Chair, member in Steering, Advisory or International Program Committees and Keynote Speaker. Now she is working in a national project with EU funding support: BioCell-NanoART ¼ Novel Bioinspired Cellular Nano-Architectures - For Digital Integrated Circuits, 3M Euro from National Authority for Scientific Research and Innovation. She is a member of European Society for Fuzzy Logic and Technology (EUSFLAT), member of Society for Industrial and Applied Mathematics (SIAM) and a Senior Member IEEE, member in Technical Committee e Fuzzy Systems (IEEE Computational Intelligence Society), chair of the Task Force 14 in Technical Committee e Emergent Technologies (IEEE CIS), member in Technical Committee e Soft Computing (IEEE SMCS). Dr. Balas was past Vice-president (responsible with Awards) of IFSA - International Fuzzy Systems Association Council (2013-2015), is a Joint Secretary of the Governing Council of Forum for Interdisciplinary Mathematics (FIM), - A Multidisciplinary Academic Body, India and recipient of the “Tudor Tanasescu” Prize from the Romanian Academy for contributions in the field of soft computing methods (2019).
xiv
About the editors
Dr. Raghvendra Kumar is working as an Associate Professor in Computer Science and Engineering Department at GIET University, India. He received his BTech, MTech, and PhD in Computer Science and Engineering, India, and Postdoc Fellow from Institute of Information Technology, Virtual Reality and Multimedia, Vietnam. He serves as a Series Editor of Internet of Everything: Security and Privacy Paradigm, Green Engineering and Technology: Concepts and Applications, published by CRC press and Taylor & Francis Group, USA, and Bio-Medical Engineering: Techniques and Applications, published by Apple Academic Press, CRC Press, and Taylor & Francis Group, USA. He also serves as an acquisition editor for Computer Science by Apple Academic Press, CRC Press, and Taylor & Francis Group, USA. He has published a number of research papers in international journal (SCI/SCIE/ESCI/Scopus) and conferences, including IEEE and Springer. He has also served as the organizing chair (RICE 2019, 2020), volume editor (RICE 2018), keynote speaker, session chair, cochair, publicity chair, publication chair, advisory board, technical program committee member in many international and national conferences, and guest editor in many special issues from reputed journals (indexed by Scopus, ESCI, SCI). He also published 13 chapters in edited book published by IGI Global, Springer, and Elsevier. His researches areas are computer networks, data mining, cloud computing and secure multiparty computations, theory of computer science, and design of algorithms. He has authored and edited 23 computer science books in field of Internet of things, data mining, biomedical engineering, big data, and robotics published by IGI Global Publication, USA, IOS Press, Netherlands, Springer, Elsevier, and CRC Press, USA. Dr. Brojo Kishore Mishra is currently working as a Professor in the Department of Computer Science and Engineering at the GIET University, Gunupur-765022, India. He received his PhD degree in Computer Science from the Berhampur University in 2012. He has published more than 30 research papers in national and international conference proceedings, 25 research papers in peer-reviewed journals, and 22 book chapters; authored 2 books; and edited 4 books. His research interests include data mining, machine learning, soft computing, and security. He has organized and co-organized local and international conferences and also edited several special issues for journals. He is the Senior Member of IEEE and Life Member of CSI, ISTE. He is the Editor of CSI Journal of Computing.
Preface Deep learning has been rapidly developed in the recent years, in terms of both methodological development and practical applications. It provides computational models of multiple processing layers to learn and represent data with multiple levels of abstraction. It is able to implicitly capture intricate structures of large-scale data and is ideally suited to some of the hardware architectures that are currently available. The purpose of this book is to provide a diverse, but complementary, set of contributions to demonstrate new developments and applications of deep learning and computational machine learning to solve problems in biomedical engineering. The proposed book will be organized as a reference source for enabling readers to have an idea about the relation between deep learning and biomedical engineering. In Chapter 1, survey of deep learning is used for image classification, carotid ultrasound data investigation, cardiotocography, intravascular ultrasound report, lung CT report, brain tumor prediction, object detection, segmentation, breast cancer prediction, ECG (electrocardiogram) signal, EEG (electroencephalogram), PPG signal registration, and psoriasis skin disease as well as cancer detection. Concise summaries are delivered of trainings per application zone: pulmonary, musculoskeletal neuro, digital pathology, and abdominal, retinal, breast, and cardiac. There are various types of deep learning techniques present to improve accuracy of the medical dataset. Chapter 2 provides a detailed discussion on different convolutional neural network (CNN) architectures and their applications in the medical imaging domain. Moreover, a stateof-the-art comparison has been carried out between several existing works inside medical imaging based on CNN. Lastly, the work concludes with several critical remarks highlighting future challenges and their solutions. Chapter 3 discusses about the class of tools that enable deep learning engineers to actually do their work faster and more effectively. Some of the tools include TensorFlow, Keras, Caffe, and Torch. Deep learning models make use of several kinds of
xvi
Preface
advanced algorithms. Some algorithms are best suited to perform specific tasks. In order to choose the right ones, it is good to gain grasp of all primary algorithms. Excellent knowledge of advanced deep learning techniques, their types, and applications can help users execute them for various purposes. Chapter 4 aims at critically analyzing the existing techniques used in blockchains and their suitability in education domain. The advantages and challenges in using blockchain-based applications in education are also discussed in this chapter. Security breaches and attacks on using blockchains are discussed along with possible countermeasures. A plan of how existing models can be improved to enhance performance of blockchains in applications belonging to education is also discussed. Chapter 5 investigates the use of different deep neural network architectures and natural language processing for depression detection in cancer communities. Depression detection using sentiment affect can be of great assistance to the doctors treating cancer patients and aid them in deciding whether along with the cancer treatment their patients need help from psychologists or psychiatrists. Chapter 6 focuses on early disease recognition that requires high-resolution images. After a preprocessing step using a contrast enhancement method, all the diseased blobs are segmented for the whole dataset. A list of several measurementbased features that represent the blobs is chosen and selected based on principle component analysis. The features are used as inputs for a standard feedforward neural network. Our results show competitive classification results not only with other deep learning approaches, such as CNNs, but also with a simpler network structure. Chapter 7 determines how deep learning helps in the early diagnosis of several diseases such as Alzheimer’s disease, rheumatic diseases, autism spectrum disorder, and more. After expanding upon the basics of deep learning and biomedical engineering, the chapter explores more upon diagnostics using deep learning and discusses the early diagnosis of certain diseases. Chapter 8 details on the advancement in the subset of machine learning; the deep learning made this research area into high potential in terms of precise prediction and accuracy. Many versions of deep learningebased architecture are implemented along with various nonvisualization and visualization techniques
Preface
to classify and detect the symptoms of plant disease with several performance metrics. This chapter illustrates a comprehensive review of deep learning models used to detect plant diseases, and in some cases, the diseases have been identified before the symptoms appear clearly. Chapter 9 discusses about fundamentals of biomedical engineering and deep learning. It also explores about applications of deep learning in various problems of biomedical field. Chapter 10 discusses the fundamentals of image processing first, including segmentation and edge detection, followed by identifying critical areas in biomedical images, denoising, and applying image processing technique on various available biomedical image datasets. The authors focus on tissue segmentation, application of CNN in interpreting biomedical images, usage of different deep learning libraries for identifying areas of interest in a biomedical image, computer-aided disease diagnosis or prognosis, and so on. We will conclude by raising research issues and suggesting future directions for further improvements. The aim of this book is to support the computational studies at the research and postgraduation level with open problemsolving techniques. We are confident that it will bridge the gap for them by supporting novel solution in their problem solving. At the end, editors have taken utmost care while finalizing the chapters to the book, but we are open to receive your constructive feedback, which will enable us to carry out necessary points in our forthcoming books. Valentina Emilia Balas Brojo Kishore Mishra Raghvendra Kumar
xvii
Key features 1. Covers the evolution of deep learning in biomedical engineering and healthcare from fundamental theories to present forms 2. Presents diversified medical applications of deep learning with use cases 3. Includes contributors from different parts of the world 4. Explores deep learning and machine learning techniques along with biomedical engineering applications 5. Presents from multiple perspectives such as academics, industry, and research fields 6. Emphasizes on the advancements and cutting-edge technologies throughout 7. Focuses on different tools, platforms, and techniques
About the book Deep learning (DL) is a method of machine learning, as running over artificial neural networks, which has a structure above the standards to deal with large amounts of data. This is generally because of the increasing amount of data, input data sizes, and, of course, greater complexity of objective real-world problems. Research studies performed in the associated literature show that the DL currently has a good performance among considered problems and seems to be a strong solution for more advanced problems of the future. In this context, this book aims to provide some essential information about DL and its applications within the field of biomedical engineering. Due to numerous biomedical information sensing devices, such as computed tomography, magnetic resonance imaging, ultrasound, single photon emission computed tomography, positron emission tomography, magnetic particle imaging, EE/MEG, optical microscopy and tomography, photoacoustic tomography, electron tomography, and atomic force microscopy, large amount of biomedical information was gathered these years. This poses a great challenge on how to develop new advanced imaging methods and computational models for efficient data processing, analysis, and modeling in clinical applications and in understanding the underlying biological process.
1 Congruence of deep learning in biomedical engineering: future prospects and challenges Aradhana Behura Veer Surendra Sai University of Technology, Burla, Sambalpur, Odisha, India
1. Introduction Death from breast cancer [7,8,11,14,34,35] may be avoided by detecting the risks of medical patients and discussing them effectively [3]. One of the known risks for tumor development besides age, gender, gene mutations, and family history is the comparative sum of radiodense tissue in the female breast, called mammographic density [26e30]. By using a stacked autoencoder we can more accurately predict brain tumors. C-means, K-means, and DBSCAN clustering techniques are used to detect affected areas in medical images. There are various types of nature-inspired algorithms used to optimize the performance of clustering that provide better results. Segmentation of brain [31] and liver tumors provides [9,10,12,13,23] important biomarkers for medical diagnosis [24,25]. Here, we present and authenticate a procedure to integrate an improved edge pointer and derive an initial curve for magnetic resonance imaging (MRI)-based disease segmentation from the dataset. At the preprocessing step, the computed tomography (CT) image intensity values were truncated to lie in a fixed range to enhance the image contrast surrounding the organ and the disease-affected area. To eliminate nonliver tissues for the following segmentation of the disease, the liver is segmented by two convolutional neural networks (CNNs) [15,21e23] in a coarse-to-fine manner. Here, we present a new procedure for combining high-resolution photorealistic medical images from the semantic label charts with conditional generative adversarial networks (GANs). A GAN [32,33] is a type of deep learning method made up of two neural networks in conflict with each other in a zero-sum game outline. Handbook of Deep Learning in Biomedical Engineering. https://doi.org/10.1016/B978-0-12-823014-5.00003-X Copyright © 2021 Elsevier Inc. All rights reserved.
1
2
Chapter 1 Congruence of deep learning in biomedical engineering
The combination of two neural networks that make up an architecture of a GAN are: • a generator with the objective of producing new examples of a thing, which will be indistinguishable from real ones, and • a discriminator with an area that classifies the duplicate (whether the particular part of the organ is affected or not in a disease). This architecture can be used in text, images, video, and audio. There are various types of CNN, which are described in [33] (Table 1.1). Currently, coronavirus (COVID-19) is a fatal disease. By using deep learning we can predict the rate of the disease and which area is affected most. From Fig. 1.1 [36], we show the artificial intelligence (AI) procedure used for coronavirus. Fig. 1.2 introduces the procedure of training and testing of data. Figs. 1.3, 1.5 and 1.6, describe the SqueezeNet and Fire model architecture, and Fig. 1.4 shows the GAN architecture. Sections 2 and 3 introduce the encryption as well as decryption of medical images to preserve authenticity. For this reason, no one can alter the patient’s personal data, which may compromise the patient’s medical information. In today’s world, the movement of information utilizing the Internet is developing quickly. Thus many users can transfer business reports and significant data, for example, by utilizing the web. Security is a significant issue when transferring information utilizing the web because unapproved individuals can hack into the information for various reasons. In data storage, cryptography and steganography are the most utilized methods for sending delicate and private data safely. An exceptional mainstream procedure to secure significant data over the Internet is the cryptography technique. In this strat-
Table 1.1 Types of convolutional neural network (CNN).
Year
CNN
Developed by
1998 2012
LeNet(8) AlexNet(7)
2013 2014 2014
ZFNet(8) GoogLeNet(l9) VGG Net(16)
Yann LeCun et al. Alex Krizhevsky, Geoffrey Hinton, and Ilya Sutskever Matthew Zeiler and Rob Fergus Google Karen Simonyan and Andrew Zisserman
Place
Top 5 error No. of rate parameters
1st
15.3%
60,000 60 million
1st 1st 2nd
14.8% 6.67% 7.3%
4 million 138 million
Chapter 1 Congruence of deep learning in biomedical engineering
Patient with symptoms
Conventional approach (non-AI)
AI based approach
Physician analyze the symptoms
Physician identify the possible match of COVID-19 symptom with AI support
If multiple matches found
If no multiple matches found
Test sample taken
No test sample taken
Samples taken to confirm infection & decide further therapy
Patient gets quarantined/admitted
Start AI based treatment & monitoring
Patients get quarantined/admitted
Symptomatic treatment started
Recovery Phase Recovery Phase
Retest for COVID-19 Retest for COVID-19
Positive
Isolation
Negative Positive
Negative
Isolation
Cured
Cured
Figure 1.1 General procedure of artificial intelligence (AI) and non-AI-based applications that help general physicians to identify COVID-19 symptoms. From Vaishya, R. et al. Artificial Intelligence (AI) applications for COVID-19 pandemic,Diabetes & Metabolic Syndrome: Clinical Research & Reviews, Volume 14, Issue 4, 2020, Pages 337e339.
egy the information takes on a structure that can only be perceived by the proposed beneficiary. Because the coded information is in an unrecognized structure, it is open to the possibility of attack. Another security strategy, i.e., data cover-up, is likewise a generally utilized method that discourages the
3
4
Chapter 1 Congruence of deep learning in biomedical engineering
Training Training Dataset
Machine Learning
Statistical Models
Prediction & Testing Historical Data
Random Sampling Testing Dataset
Model Validation Outcome
Prediction
New Data
Model
Predicted Outcome
Figure 1.2 Training and prediction scheme.
Figure 1.3 SqueezeNet architecture [33]. Adapted from sources: https://medium.com/analytics-vidhya/; Amit Kumar, Rama Chellappa, A Convolution Tree with Deconvolution Branches: Exploiting Geometric Relationships for Single Shot Keypoint Detection, arXiv:1704.01880 [cs.CV].
aggressor by disguising the data inside the bearer. This approach provides higher security and can guarantee message delivery. Steganography is the practice of hiding information within other less secret information. It is a Greek word: stegano infers covered or concealed and graphy infers writing. There are many ways to hide information, for instance, propelled pictures, chronicles, sound reports, and other PC records; however, modernized pictures are the most notable.
Chapter 1 Congruence of deep learning in biomedical engineering
Figure 1.4 Generative adversarial network architecture. Based on from the sources: https://towardsdatascience.com/ review-squeezenet-image-classification-e7414825581a; Amit Kumar, Rama Chellappa, A Convolution Tree with Deconvolution Branches: Exploiting Geometric Relationships for Single Shot Keypoint Detection, arXiv:1704.01880 [cs.CV].
The procedure of stego image embedding as well as the extraction procedure is described in Fig. 1.11 by using an autoencoder, which is a deep learning technique. We can also use other deep learning models like CNN, GAN, etc. By using these types of image encryption as well as decryption methods, hospitals/physicians can preserve a patient’s information in the patient report, MRI, and CT scan data.
5
6
Chapter 1 Congruence of deep learning in biomedical engineering
Figure 1.5 Fire module with the hyperparameters: s1 1 ¼ 3, e1 1 ¼ 4, and e3 3 ¼ 4 [37]. From Forrest N. Iandola, Matthew W. Moskewicz, Khalid Ashraf, Song Han, William J. Dally and Kurt Keutzer, SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and